CN109495744A - The big multiplying power remote sensing image compression method of confrontation network is generated based on joint - Google Patents
The big multiplying power remote sensing image compression method of confrontation network is generated based on joint Download PDFInfo
- Publication number
- CN109495744A CN109495744A CN201811270154.7A CN201811270154A CN109495744A CN 109495744 A CN109495744 A CN 109495744A CN 201811270154 A CN201811270154 A CN 201811270154A CN 109495744 A CN109495744 A CN 109495744A
- Authority
- CN
- China
- Prior art keywords
- network
- remote sensing
- sensing image
- joint
- confrontation
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/10—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding
- H04N19/102—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the element, parameter or selection affected or controlled by the adaptive coding
- H04N19/124—Quantisation
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/10—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding
- H04N19/102—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the element, parameter or selection affected or controlled by the adaptive coding
- H04N19/13—Adaptive entropy coding, e.g. adaptive variable length coding [AVLC] or context adaptive binary arithmetic coding [CABAC]
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Signal Processing (AREA)
- Compression, Expansion, Code Conversion, And Decoders (AREA)
- Compression Of Band Width Or Redundancy In Fax (AREA)
Abstract
The invention discloses a kind of big multiplying power remote sensing image compression methods based on more structural generations confrontation network, solve conventional method and do not excavate the technical issues of correlation and big multiplying power compress lower eye fidelity decline between image pixel.It realizes that step has: remote sensing image data collection is pre-processed；Pretreated remote sensing image data collection input joint generates confrontation network training and obtains trained network model；Big multiplying power remote sensing image compression, which is carried out, with the network that training obtains obtains coding file；Entropy coding is done to network code result and obtains ASCII stream file ASCII to be transmitted；Original remote sensing image is obtained with the decompression that decoding network carries out ASCII stream file ASCII.The present invention generates confrontation network remote sensing image using joint and establishes the big multiplying power compression of model progress, encoding and decoding network symmetrical structure is carried out dual training by the network respectively, by training so that model extracts the correlation between the correlation of image itself and remote sensing image simultaneously, improves big multiplying power and compress lower eye fidelity.
Description
Technical field
The invention belongs to image processing technique fields, are related to the big multiplying power compression of image, specifically a kind of based on generation pair
The big multiplying power compression method of the remote sensing image of anti-network is used for remote sensing image compression.
Background technique
With the increase of in-orbit remote sensing satellite number and the increase of imaging resolution, the remote sensing image data acquired on star is got over
Come more, the demand to more high magnification data compression technique is increasingly urgent to.Develop the big multiplying power compress technique of remote sensing image, favorably
In saving channel capacity, pass data on star back ground in time under downstream transmission bandwidth limitation, image this in national defence, social section
The fields such as, environmental protection, urban planning important role.
Currently, the big multiplying power compression method of remote sensing image image mainly uses JPEG2000 method, this method is based on small
The image compression standard of wave conversion organizes creation and maintenance by Joint Photographic Experts Group, obtains at present almost best big times
Rate compression effectiveness.But since there are potential sparsities in more physical attributes of high-resolution satellite remote sensing date, so remote sensing
Many difficult points that image compression has it intrinsic: remote sensing image scale is big, and redundancy is strong, and demand is compressed offline, and stability is strong, greatly
Multiplying power etc., so that also seldom for remote sensing image compression technology made by remote sensing image, big multiplying power remote sensing image compression technology is more
It is very few.
JPEG2000 realizes the big multiplying power compression of remote sensing image, but its method is just with the correlation of image itself
Property, the redundancy without reducing cataloged procedure using correlation between this kind of image of remote sensing image.Moreover, JPEG2000 is compressing
When multiplying power is very big, subjective performance, such as visual information fidelity are poor.
Summary of the invention
The shortcomings that it is an object of the invention to overcome above-mentioned prior art and deficiency propose that one kind makes full use of remote sensing image
Correlation joint generates the big multiplying power remote sensing image compression method of confrontation network between pixel, and this method compresses Shi Nengda in big multiplying power
To better subjective effect.
The present invention is a kind of big multiplying power remote sensing image compression method that confrontation network is generated based on joint, which is characterized in that
It comprises the following steps that
(1) pre-process to original remote sensing image data collection: input needs the remote sensing image compressed, by normalizing
Change, sliding window cutting obtains data set to be trained；
(2) building joint generates confrontation network: first layer to the layer 5 that joint generates confrontation network is coding network, is used
In the coding to remote sensing image；Layer 6 to the eleventh floor that joint generates confrontation network is decoding network, for remote sensing shadow
The decoding of picture；Wherein first layer to the 4th layer respectively with eleventh floor to the identical pattern image of the corresponding size of layer 7 at
Two inputs of joint confrontation network, joint confrontation network is together constituted with coding network, decoding network combines generation confrontation net
Network；
(3) joint generation confrontation network is inputted to pretreated remote sensing image data collection to be trained, trained
Network model: the network model successively includes coding network, quantizer, decoding network, fights four parts of network；
(4) coding network in confrontation network is generated with the joint that training obtains and carry out big multiplying power remote sensing image compression: defeated
Enter original remote sensing image to be compressed, after pretreatment, inputs in trained model, by coding network and quantizer
Obtain the coding result that joint generates confrontation network；
(5) distich symphysis carries out lossless entropy coding at confrontation network code result, obtains ASCII stream file ASCII to be transmitted；
(6) solution for the ASCII stream file ASCII that the decoding network in confrontation network is transmitted is generated with the joint that training obtains
Compression: the ASCII stream file ASCII received is input in decoding network, the remote sensing image restored after being compressed.
Present invention application generates confrontation network and extracts the correlation between remote sensing image to improve compression ratio, saves storage and passes
Defeated space.
Compared with the prior art, the present invention has the following advantages:
A. joint proposed by the present invention generates confrontation network and encoding and decoding network symmetrical structure is carried out dual training respectively, makes
Encoding and decoding network tend to be reciprocal in training, be conducive to improve network overfitting problem, by training so that model simultaneously
The correlation between the correlation and remote sensing image of image itself is extracted, big multiplying power is improved and compresses lower eye fidelity.
B. confrontation model is generated present invention employs joint to be trained, biography is overcome to efficiently portraying for loss function
The limitation of Manual definition's loss function in system method, the characteristic information of remote sensing image is made full use of, and is obtained more efficient
Compact model.
C. present invention uses the residual error network modules with powerful characteristic present ability as coding network and decoding net
The composed structure of network, complicated structure improves the ability of model extraction feature, thus between taking full advantage of remote sensing image
Correlation.
D. the experimental results showed that, the present invention on specific remote sensing image compared with the compression methods such as JPEG2000, big times
Visual information fidelity is higher when rate is compressed.
Detailed description of the invention
Fig. 1 is flow chart of the invention；
Fig. 2 is under 256 multiplying powers of the invention, and the compression reconfiguration to high score image is as a result, it is the distant of num1 that Fig. 2 a, which is number,
Feel image original, Fig. 2 b is the result figure that Fig. 2 a compression result is reconstructed with method of the invention, and Fig. 2 c is to use
The result figure that Fig. 2 a compression result is reconstructed in the method for JPEG2000；
Fig. 3 is under 256 multiplying powers of the invention, and the compression reconfiguration to high score image is as a result, it is the distant of num2 that Fig. 3 a, which is number,
Feel image original, Fig. 3 b is the result figure that Fig. 3 a compression result is reconstructed with method of the invention, and Fig. 3 c is to use
The result figure that Fig. 3 a compression result is reconstructed in the method for JPEG2000；
Fig. 4 is under 256 multiplying powers of the invention, and the compression reconfiguration to high score image is as a result, it is the distant of num3 that Fig. 4 a, which is number,
Feel image original, Fig. 4 b is the result figure that Fig. 4 a compression result is reconstructed with method of the invention, and Fig. 4 c is to use
The result figure that Fig. 4 a compression result is reconstructed in the method for JPEG2000；
Fig. 5 is residual error structure chart used in the present invention.
Specific embodiment
With reference to the accompanying drawing with example to the detailed description of the invention:
Embodiment 1
Currently, remote sensing image compression is in the Disciplinary Frontiers of remote sensing technology, put into both at home and abroad a large amount of manpower and
Material resources studies it, but also faces huge difficulty in terms of the real-time Transmission of remote sensing image data, for this
The one problem present invention is unfolded to study from the angle of image, has more efficient big multiplying power compression method in mind, proposes to generate based on joint
Fight the big multiplying power compression method of remote sensing image of network.
The present invention is a kind of big multiplying power remote sensing image compression method that confrontation network is generated based on joint, referring to Fig. 1, including
It has the following steps:
(1) pre-process to original remote sensing image data collection: input needs the remote sensing image compressed, by normalizing
Change, sliding window cutting obtains data set to be trained.Obtain pretreated remote sensing image data collection；Remote sensing image general size is all
It is very big, due to calculating the limitation of power, image cutting is needed to be input in network again later.
(2) building joint generates confrontation network: joint of the invention generates confrontation network towards remote sensing image compression, referring to
Fig. 1, first layer to the layer 5 that joint generates confrontation network is coding network, for the coding to remote sensing image；Joint generates
Layer 6 to the eleventh floor for fighting network is decoding network, for the decoding to remote sensing image；Wherein first layer is to the 4th layer
Respectively with eleventh floor to the identical pattern image of the corresponding size of layer 7 at two inputs of joint confrontation network, joint
Confrontation network is together constituted with coding network, decoding network combines generation confrontation network, and coding network is inverse each other with decoding network
Process, then combine and be gradually intended to reversible structure under the training mode for generating confrontation network.
(3) joint generation confrontation network is inputted to pretreated remote sensing image data collection to be trained, trained
Network model.Referring to Fig. 1, which successively includes coding network, quantizer, decoding network, confrontation four portions of network
Point；Wherein coding network, quantizer and decoding network constitute compact model, and confrontation network is for correction model parameter in training.
Referring to Fig. 1, the present invention generates confrontation network remote sensing image using joint and establishes model and compress, and the network is by encoding and decoding network
Symmetrical structure carries out dual training respectively, so that encoding and decoding network tends to be reciprocal in training, advantageously accounts for the excessively quasi- of network
Conjunction problem is improved by training so that model extracts the correlation between the correlation of image itself and remote sensing image simultaneously
Big multiplying power compresses lower eye fidelity.
(4) coding network in confrontation network is generated with the joint that training obtains and carry out big multiplying power remote sensing image compression: defeated
Enter original remote sensing image to be compressed, after the pretreatment in step (1), in input step (3) in trained model,
The coding result that joint generates confrontation network is obtained by coding network and quantizer；The size of the coding result corresponds to original image
Size form fixed compression bit rate.
(5) distich symphysis carries out lossless entropy coding at confrontation network code result, and remote sensing image is to be transmitted after being compressed
ASCII stream file ASCII.Distich symphysis uses the lossless entropy coding of Huffman at confrontation network code result in this example.In addition to Huffman without
Entropy coding is damaged, the present invention can also realize lossless entropy coding with arithmetic coding.
(6) code that the decoding network in confrontation network is transmitted is generated with the joint that training in step (3) obtains
The decompression of stream file: the ASCII stream file ASCII received is input in decoding network, is completed big multiplying power compression process, is obtained remote sensing shadow
As the remote sensing image restored after compression.
The present invention is compressed by establishing model to remote sensing image data collection, based on deep learning, between image
Correlation improve compression accuracy with the mode that the correlation of image itself combines, while by using lossless entropy coding into
The second-compressed of row coding, improves compression efficiency.
Embodiment 2
The big multiplying power compression method of remote sensing image of confrontation network is generated with embodiment 1 based on joint, wherein institute in step (1)
That states pre-processes original remote sensing image data collection, comprises the following steps that
(1a) carries out sliding window cutting to original remote sensing image and forms training dataset；
The data set after cutting is normalized in (1b): obtaining pretreated remote sensing image data collection；Normalize letter
Number is
Wherein, x*Indicate the value after normalization, min indicates the minimum value of image pixel, and max indicates image pixel most
Big value.
The present invention expands training dataset by the way of sliding window cutting, by the data set of production multiplicity, promotes compression
The generalization ability of model.
Embodiment 3
The big multiplying power compression method of remote sensing image of confrontation network is generated with embodiment 1-2 based on joint, wherein step (3) institute
That states is trained pretreated remote sensing image data collection input joint generation confrontation network, specifically includes following steps
It includes coding network, magnitude device, decoding network and more structures confrontation network four that (3a) joint, which generates confrontation network,
A part；Wherein coding network, quantizer and decoding network constitute compact model, joint confrontation network in training with decoding net
Network carries out generating the parameter fought to adjust coding network and decoding network；Loss function in training are as follows:
Wherein, G represents coding network and decoding network, DiThe confrontation network in code each stage, f (y)=(y-1)2, g
(y)=y2, d (x, y) represents least square method error vector.
(3b) during trained each step, in addition to fight loss other than, be additionally added least square method calculate error to
Amount updates weight according to standard back-propagation algorithm.
The characteristic pattern point of size is corresponded in (3c) decoding network in the characteristic pattern and corresponding decoding network of generation at all levels
It Shu Ru not fight in network individually, form joint confrontation network.The target of each confrontation network is to maximize the maximum in the stage
Metaplasia at the difference of image and original image, be responsible for minimizing the difference for generating image and original image by coding network and decoding network
Not.
The present invention has more efficient learning efficiency compared to conventional method, and training 10 times uses the dynamical learning rate side Adam
Method adjusts the learning rate of each parameter using the single order moments estimation of gradient and second order moments estimation dynamic.The advantages of Adam, is main
It is after bias correction, iterative learning rate has a determining range each time, so that parameter is more steady.The present invention is instructing
During experienced each step, other than fighting loss, it is additionally added least square method and calculates error vector, reversely passed according to standard
It broadcasts algorithm and updates weight, increase the ability of model extraction feature, avoid the simple inefficient study of conventional method.
Embodiment 4
The big multiplying power compression method of remote sensing image of confrontation network is generated with embodiment 1-3, the coding net based on joint
Network: referring to Fig. 1, convolutional layer therein uses five convolutional layers, the filter that each convolutional layer step number is 2, activation primitive
Amendment linear function (Relu) is selected, residual error network structure is added in the last layer of convolutional layer.
Present invention uses the residual error network module with outstanding feature extraction functions as coding network one of them
Unit, referring to Fig. 3, complicated structure improves the ability of model extraction feature, takes full advantage of the correlation in remote sensing image
Information.Step number is used to improve the ability of model extraction feature instead of traditional down-sampling layer for 2 filter in this example, thus
Improve the compression efficiency of model.
Embodiment 5
Hyperspectral imaging lossless compression method based on deep learning is with embodiment 1-4, quantizer in the present invention: quantization
Device uses softmax layers, quantifies to each pixel.
The present invention carries out quantizing process with softmax layers of classification, both can solve conventional quantization method non-differentiability in this way
Problem, and crucial connection function can be played in network training, allows the training that network is end-to-end.
Embodiment 6
The big multiplying power compression method of remote sensing image of confrontation network is generated with embodiment 1-5, decoding of the invention based on joint
Network uses five layers of warp lamination, and referring to Fig. 1, the first layer of the filter that each convolutional layer step number is 2, convolutional layer is added
Residual error network structure.Decoding network and coding network are at reciprocal structure.
After decoding, the entirely big multiplying power whole compression process based on remote sensing image is completed.
Natural image compression field presents its development potentiality again for current depth study, and the present invention passes through depth
The combination of habit technology and extensive remote sensing image, building depth network can be very good to realize that the big multiplying power compression of remote sensing image needs
It asks.Remote sensing technology is compressed using depth learning technology, is conducive to extract the redundancy feature inside remote sensing image, favorably
In saving bandwidth capacity, the development bottleneck of aerospace industry remote sensing image temporal resolution, spatial resolution is solved.
Technical effect of the invention is explained again below by experimental data:
Embodiment 7
The big multiplying power compression method of remote sensing image of confrontation network is generated with embodiment 1-6 based on joint, remote sensing in the present invention
The big multiplying power compression of image uses 2 number of high score, obtains 10000 trained pictures and 300 tests by cutting data extending
Picture is tested under the Tensorflow experiment porch for the server for having two pieces of 1080 video cards of GTX.This is listed in table 1
The effect that the compression method of invention and the compression method of JPEG2000 respectively compress 2 number of high score compares, wherein pressing
Contracting effect is concerned with this four indexs using visual information fidelity, structural similarity, Y-PSNR, relative polarities edge come table
Show.
Table 1 compares 256 multiplying power compression effectiveness of No. 2 remote sensing images of high score,
From the comparison of above-mentioned experimental result, it can be seen that the present invention opens the compression of remote sensing image to num1-num3 tri-
Effect increases, and visual information fidelity, relative polarities edge the two indexs that are concerned with are above same multiplying
JPEG2000, structural similarity and Y-PSNR are similar to the result of JPEG2000 method.Wherein, the structure of num2 is similar
Property index has also exceeded JPEG2000.
Present invention employs the modes for generating confrontation to be trained, and efficient gradient transmitting overcomes traditional method for extracting
The limitation of information makes full use of the data information between remote sensing image, obtains accurate result.
In brief, the big multiplying power remote sensing image compression method disclosed by the invention that confrontation network is generated based on joint, is belonged to
In image compression field.Solving conventional method does not utilize correlation between remote sensing image, and big multiplying power to compress lower visual fidelity
The problems such as degree is decreased obviously.It realizes that step includes: to pre-process to original remote sensing image data collection: inputting to be compressed distant
Feel image, obtains pretreated remote sensing image；Building joint generates confrontation network: joint generates the first layer of confrontation network extremely
Layer 5 is coding network, for the coding to remote sensing image；Layer 6 to the eleventh floor that joint generates confrontation network is solution
Code network, for the decoding to remote sensing image；Wherein first layer is corresponding to layer 7 with eleventh floor respectively to the 4th layer
The identical pattern image of size combines confrontation network and coding network, decoding network one at two inputs of joint confrontation network
With constitute joint generate confrontation network to pretreated remote sensing image data collection input joint generate confrontation network be trained,
Obtain trained network model: the network includes coding network, binaryzation device, decoding network, confrontation four parts of network；With
The network that training obtains carries out big multiplying power remote sensing image compression: original remote sensing image to be compressed is inputted, after pretreatment,
It inputs in trained model, obtains ASCII stream file ASCII to be transmitted by coding network and binaryzation device；To network code result
Entropy coding is carried out, ASCII stream file ASCII to be transmitted is obtained；The decompression of ASCII stream file ASCII is carried out with the network that training obtains: by what is received
ASCII stream file ASCII is input in decoding network, the remote sensing image restored.The present invention generates confrontation network remote sensing shadow using joint
It is compressed as establishing model, which carries out dual training for encoding and decoding network symmetrical structure respectively, so that encoding and decoding network
Tend to be reciprocal in training, advantageously account for the overfitting problem of network, by training so that model extracts image itself simultaneously
Correlation and remote sensing image between correlation, improve big multiplying power and compress lower eye fidelity.It is led applied to image compression
Domain.
Claims (6)
1. a kind of big multiplying power remote sensing image compression method for generating confrontation network based on joint, which is characterized in that include as follows
Step:
(1) pre-process to original remote sensing image data collection: input needs the remote sensing image compressed, by normalizing, sliding
Window cutting obtains data set to be trained；
(2) building joint generate confrontation network: joint generate confrontation network first layer to layer 5 be coding network, for pair
The coding of remote sensing image；Layer 6 to the eleventh floor that joint generates confrontation network is decoding network, for remote sensing image
Decoding；Wherein first layer to the 4th layer respectively with eleventh floor to the identical pattern image of the corresponding size of layer 7 at joint
Two inputs of network are fought, joint confrontation network is together constituted with coding network, decoding network combines generation confrontation network；
(3) joint generation confrontation network is inputted to pretreated remote sensing image data collection to be trained, obtain trained net
Network model: the network model successively includes coding network, quantizer, decoding network, confrontation four parts of network；
(4) generate the coding network in confrontation network with the joint that training obtains and carry out big multiplying power remote sensing image compression: input is former
The remote sensing image to be compressed to begin is inputted in trained model, is obtained by coding network and quantizer after pretreatment
Joint generates the coding result of confrontation network；
(5) distich symphysis carries out lossless entropy coding at confrontation network code result, obtains ASCII stream file ASCII to be transmitted；
(6) decompression for the ASCII stream file ASCII that the decoding network in confrontation network is transmitted is generated with the joint that training obtains
Contracting: the ASCII stream file ASCII received is input in decoding network, the remote sensing image restored.
2. the big multiplying power remote sensing image compression method according to claim 1 for generating confrontation network based on joint, feature
It is, wherein original remote sensing image data collection is pre-processed described in step (1), is comprised the following steps that
(1a) carries out sliding window cutting to original remote sensing image and forms training dataset；
The data set after cutting is normalized in (1b), obtains pretreated remote sensing image data collection.
3. the big multiplying power remote sensing image compression method according to claim 1 for generating confrontation network based on joint, feature
It is, joint generation confrontation network wherein is inputted to pretreated remote sensing image data collection described in step (3) and is trained,
Specifically include following steps
It includes coding network, magnitude device, decoding network and more structures confrontation four portions of network that (3a) joint, which generates confrontation network,
Point；Wherein coding network, quantizer and decoding network constitute compact model, joint confrontation network in training with decoding network into
Row generates the parameter fought to adjust coding network and decoding network；
(3b) other than fighting loss, is additionally added least square method and calculates error vector, root during trained each step
Weight is updated according to standard back-propagation algorithm；
The characteristic pattern difference for corresponding to size in (3c) decoding network in the characteristic pattern and corresponding decoding network of generation at all levels is defeated
Enter in single confrontation network, forms more structure confrontation networks；The target of each confrontation network is to maximize the maximization in the stage
The difference of image and original image is generated, coding network and decoding network are responsible for minimizing the difference for generating image and original image
Not.
4. the big multiplying power remote sensing image compression method of confrontation network is generated based on joint according to claim 1 or described in 3,
Be characterized in that, the coding network: convolutional layer therein uses five convolutional layers, the filter that each convolutional layer step number is 2
Wave device, activation primitive select amendment linear function (Relu), and residual error network structure is added in the last layer of convolutional layer.
5. the big multiplying power remote sensing image compression method according to claim 1 or 3 for generating confrontation network based on joint, special
Sign is that the quantizer: quantizer uses softmax layers, quantifies to each pixel.
6. the big multiplying power remote sensing image compression method according to claim 1 or 3 for generating confrontation network based on joint, special
Sign is that the decoding network: decoding network is using five layers of warp lamination, the filter that each convolutional layer step number is 2,
Residual error network structure is added in the first layer of convolutional layer.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811270154.7A CN109495744B (en) | 2018-10-29 | 2018-10-29 | Large-magnification remote sensing image compression method based on joint generation countermeasure network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811270154.7A CN109495744B (en) | 2018-10-29 | 2018-10-29 | Large-magnification remote sensing image compression method based on joint generation countermeasure network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109495744A true CN109495744A (en) | 2019-03-19 |
| CN109495744B CN109495744B (en) | 2019-12-24 |
Family
ID=65693275
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811270154.7A Active CN109495744B (en) | 2018-10-29 | 2018-10-29 | Large-magnification remote sensing image compression method based on joint generation countermeasure network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109495744B (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110062125A (en) * | 2019-04-18 | 2019-07-26 | 西安万像电子科技有限公司 | Image transfer method and device |
| CN110225350A (en) * | 2019-05-30 | 2019-09-10 | 西安电子科技大学 | Natural image compression method based on production confrontation network |
| CN110348487A (en) * | 2019-06-13 | 2019-10-18 | 武汉大学 | A kind of method for compressing high spectrum image and device based on deep learning |
| CN110909509A (en) * | 2019-11-28 | 2020-03-24 | 哈尔滨理工大学 | Bearing life prediction method based on InfoLSGAN and AC algorithm |
| CN111340901A (en) * | 2020-02-19 | 2020-06-26 | 国网浙江省电力有限公司 | Compression method of power transmission network picture in complex environment based on generating type countermeasure network |
| CN113470127A (en) * | 2021-09-06 | 2021-10-01 | 成都国星宇航科技有限公司 | Optical image effective compression method based on satellite-borne cloud detection |
| CN116112694A (en) * | 2022-12-09 | 2023-05-12 | 无锡天宸嘉航科技有限公司 | Video data coding method and system applied to model training |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103632671A (en) * | 2013-06-28 | 2014-03-12 | 华为软件技术有限公司 | Data encoding method, data decoding method, data encoding device, data decoding device and data communication system |
| CN106845471A (en) * | 2017-02-20 | 2017-06-13 | 深圳市唯特视科技有限公司 | A kind of vision significance Forecasting Methodology based on generation confrontation network |
| AU2018100325A4 (en) * | 2018-03-15 | 2018-04-26 | Nian, Xilai MR | A New Method For Fast Images And Videos Coloring By Using Conditional Generative Adversarial Networks |
| CN108257195A (en) * | 2018-02-23 | 2018-07-06 | 深圳市唯特视科技有限公司 | A kind of facial expression synthetic method that generation confrontation network is compared based on geometry |
| CN108399625A (en) * | 2018-02-28 | 2018-08-14 | 电子科技大学 | A kind of SAR image orientation generation method generating confrontation network based on depth convolution |
| CN108495132A (en) * | 2018-02-05 | 2018-09-04 | 西安电子科技大学 | The big multiplying power compression method of remote sensing image based on lightweight depth convolutional network |
| CN108537742A (en) * | 2018-03-09 | 2018-09-14 | 天津大学 | A kind of panchromatic sharpening method of remote sensing images based on generation confrontation network |
| CN108595916A (en) * | 2018-05-10 | 2018-09-28 | 浙江工业大学 | Estimating method is composed entirely based on the gene expression for generating confrontation network |
-
2018
- 2018-10-29 CN CN201811270154.7A patent/CN109495744B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103632671A (en) * | 2013-06-28 | 2014-03-12 | 华为软件技术有限公司 | Data encoding method, data decoding method, data encoding device, data decoding device and data communication system |
| CN106845471A (en) * | 2017-02-20 | 2017-06-13 | 深圳市唯特视科技有限公司 | A kind of vision significance Forecasting Methodology based on generation confrontation network |
| CN108495132A (en) * | 2018-02-05 | 2018-09-04 | 西安电子科技大学 | The big multiplying power compression method of remote sensing image based on lightweight depth convolutional network |
| CN108257195A (en) * | 2018-02-23 | 2018-07-06 | 深圳市唯特视科技有限公司 | A kind of facial expression synthetic method that generation confrontation network is compared based on geometry |
| CN108399625A (en) * | 2018-02-28 | 2018-08-14 | 电子科技大学 | A kind of SAR image orientation generation method generating confrontation network based on depth convolution |
| CN108537742A (en) * | 2018-03-09 | 2018-09-14 | 天津大学 | A kind of panchromatic sharpening method of remote sensing images based on generation confrontation network |
| AU2018100325A4 (en) * | 2018-03-15 | 2018-04-26 | Nian, Xilai MR | A New Method For Fast Images And Videos Coloring By Using Conditional Generative Adversarial Networks |
| CN108595916A (en) * | 2018-05-10 | 2018-09-28 | 浙江工业大学 | Estimating method is composed entirely based on the gene expression for generating confrontation network |
Non-Patent Citations (2)
| Title |
|---|
| MORTEZA MARDANI ET AL: "Recurrent generative adversarial neural networks for compressive imaging", 《2017 IEEE 7TH INTERNATIONAL WORKSHOP ON COMPUTATIONAL ADVANCES IN MULTI-SENSOR ADAPTIVE PROCESSING (CAMSAP)》 * |
| 朱清: "基于生成式对抗网络的遥感影像云检测方法", 《地理空间信息》 * |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110062125A (en) * | 2019-04-18 | 2019-07-26 | 西安万像电子科技有限公司 | Image transfer method and device |
| CN110225350A (en) * | 2019-05-30 | 2019-09-10 | 西安电子科技大学 | Natural image compression method based on production confrontation network |
| CN110225350B (en) * | 2019-05-30 | 2021-03-23 | 西安电子科技大学 | Natural image compression method based on generation type countermeasure network |
| CN110348487A (en) * | 2019-06-13 | 2019-10-18 | 武汉大学 | A kind of method for compressing high spectrum image and device based on deep learning |
| CN110348487B (en) * | 2019-06-13 | 2021-07-20 | 武汉大学 | Hyperspectral image compression method and device based on deep learning |
| CN110909509A (en) * | 2019-11-28 | 2020-03-24 | 哈尔滨理工大学 | Bearing life prediction method based on InfoLSGAN and AC algorithm |
| CN110909509B (en) * | 2019-11-28 | 2022-08-05 | 哈尔滨理工大学 | Bearing life prediction method based on InfoLSGAN and AC algorithm |
| CN111340901A (en) * | 2020-02-19 | 2020-06-26 | 国网浙江省电力有限公司 | Compression method of power transmission network picture in complex environment based on generating type countermeasure network |
| CN111340901B (en) * | 2020-02-19 | 2023-08-11 | 国网浙江省电力有限公司 | Compression method of power transmission network picture under complex environment based on generation type countermeasure network |
| CN113470127A (en) * | 2021-09-06 | 2021-10-01 | 成都国星宇航科技有限公司 | Optical image effective compression method based on satellite-borne cloud detection |
| CN116112694A (en) * | 2022-12-09 | 2023-05-12 | 无锡天宸嘉航科技有限公司 | Video data coding method and system applied to model training |
| CN116112694B (en) * | 2022-12-09 | 2023-12-15 | 无锡天宸嘉航科技有限公司 | Video data coding method and system applied to model training |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109495744B (en) | 2019-12-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109495744A (en) | The big multiplying power remote sensing image compression method of confrontation network is generated based on joint | |
| CN110517329B (en) | Deep learning image compression method based on semantic analysis | |
| CN110751597B (en) | Video super-resolution method based on coding damage repair | |
| CN112734867B (en) | Multispectral image compression method and multispectral image compression system based on spatial spectrum feature separation and extraction | |
| CN104574336B (en) | Super-resolution image reconstruction system based on adaptive sub- mould dictionary selection | |
| CN110290387A (en) | A kind of method for compressing image based on generation model | |
| CN109903351B (en) | Image compression method based on combination of convolutional neural network and traditional coding | |
| CN108960333A (en) | Lossless compression method for high spectrum image based on deep learning | |
| Dong et al. | Semantic communication system based on semantic slice models propagation | |
| CN108495132B (en) | The big multiplying power compression method of remote sensing image based on lightweight depth convolutional network | |
| CN109949222A (en) | Image super-resolution rebuilding method based on grapheme | |
| CN105721869B (en) | The collection of compression tensor and reconfiguration system based on structural sparse | |
| CN110225350A (en) | Natural image compression method based on production confrontation network | |
| Li et al. | Multiple description coding based on convolutional auto-encoder | |
| CN115955563A (en) | Satellite-ground combined multispectral remote sensing image compression method and system | |
| CN111479286B (en) | Data processing method for reducing communication flow of edge computing system | |
| CN111050170A (en) | Image compression system construction method, compression system and method based on GAN | |
| CN111080729A (en) | Method and system for constructing training picture compression network based on Attention mechanism | |
| CN114494569B (en) | Cloud rendering method and device based on lightweight neural network and residual streaming | |
| Prasetyo et al. | Integrating Companding and Deep Learning on Bandwidth-Limited Image Transmission | |
| Prantl | Image compression overview | |
| Xin et al. | Soft compression for lossless image coding | |
| KR102552308B1 (en) | Haze removal appraratus and method using unified function based on neural networks | |
| CN114501034B (en) | Image compression method and medium based on discrete Gaussian mixture super prior and Mask | |
| Teng et al. | Light Field Compression Based on Graph Sample and Aggregaet Algorithm |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |